CHOOSESecurity & safety
Global Thermonuclear War Bench
Nic Cusworth@niccusworth
This project benchmarks an AI model named JEV against four other conversational AIs, noting that JEV did not speak and withheld action on a coin flip.
Global Thermonuclear War Bench: possibly the most important AI benchmark humanity will ever run. Today: JEV, a model that cannot speak, against the four chattiest AIs we have tested. It held fire on a coin flip. @typesafeai #AI #LLM youtu.be/ShtDTQdDkvs